What problem does it solve? When a loopgen run fails, stalls, hits max iterations, or comes back green but suspicious, it is hard to tell whether the agent, the spec, or the workspace is at fault. This Skill reads the run report, classifies the failure by its outcome field, and pinpoints the root cause without spending more agent budget. ## Core Features & Use Cases - Outcome-based triage: Classifies runs by outcome (max-iterations, preflight-failed, error, baseline-vacuous, spec-tampered, aborted, vacuous-success) and maps each to a likely cause and fix. - Budget-free reproduction: Confirms root causes with lint and by running the failing check command by hand, avoiding costly agent re-runs. - False-positive detection: Treats green runs with warnings (no files changed, agent did not complete) as suspect and verifies the checks actually test the requirement. - Use Case: A loop run exhausts its 5 iterations with the same check failing. The Skill reads the last iteration's evaluator feedback and agent stopReason, reproduces the failing command in the resolved workspace, and identifies a wrong test command in the evaluator options. ## Quick Start Ask the assistant to diagnose why your .loop.yaml run failed, providing the terminal output or the --report JSON file.
